Colorectal cancer (CRC) is the fourth most common type of cancer for men and women in the United Kingdom (UK) and the second leading cause of cancer death [1]. Compared to the rest of the UK, bowel cancer incidence is highest in Scotland [2] and it is projected that the number of new CRC cases in Scotland will increase by 43% by 2023-27 compared to 2008-12 [3]. Since the introduction of the Scottish Bowel Screening Programme in June 2007, uptake of screening has continued to increase to an all-time high of 65% during the period 1st May 2019 to 30th April 2021 [4]. The programme aims to detect CRC earlier and even prevent CRC occurring in the first place by identifying and removing pre-cancerous adenomatous polyps.

Despite improvements in CRC survival over time, Scotland lags behind its UK and European counterparts [5, 6]. Previous research has shown that socio-economic deprivation and remoteness factors, for example distance from a cancer centre, are significantly associated with poorer survival in Scotland [7, 8] and in the UK [9–11]. However, there are considerable differences in conclusions across studies and the mechanisms behind the observed relationships remain unclear [12–14].

Research in this area, and ultimately patient outcomes, may be improved by utilising the vast amounts of administrative healthcare data that are collected routinely as part of the delivery of patient care [16]. These data provide an opportunity to generate evidence with a high degree of external validity, being entirely representative of current care [17].

In England, linked administrative datasets have been used to investigate routes to CRC diagnosis [18]; explore provider differences in post-colonoscopy CRC diagnosis rates [19]; explain variation in treatment and outcomes [20, 21]; and describe management of disease [22]. In Scotland, there are very few CRC studies using linked administrative datasets to investigate CRC treatment and many have used administrative data from a single geographic area e.g. [12, 14, 23]. Furthermore, published, population level statistics on colorectal cancer survival [15] group stages and/or disease sites together, without distinguishing patients who are treated at early-stage or accounting for important prognostic factors.

In this study, we contribute to the existing research by providing a contemporary, population level description of the demographic, diagnostic and treatment characteristics of patients diagnosed with CRC in Scotland. We also carry out an exploratory analysis of the factors associated with survival for those early-stage, surgically treated patients. We use a newly established, unique CRC dataset, which links demographic data to the Scottish Cancer Registry and routinely collected hospital admissions data. Full details of the linkage and creation of this source dataset are described elsewhere [24]. Public Health Scotland’s (PHS) Data Quality Assurance team carry out regular data accuracy audits on national datasets. Unlike other published work, we focus on patients defined as early-stage disease treated with surgery and incorporate prognostic factors such as disease stage and comorbidities. In what follows, we firstly describe the demographic, diagnostic and treatment characteristics of patients with colon and rectal cancer. Secondly, for those patients identified as early-stage, surgically treated, we estimate their survival and explore the factors affecting their overall survival and CRC specific survival.

The study design is a retrospective cohort study of adults with an incident CRC registered on the Scottish Cancer Registry (International Disease Classification 10th Revision (ICD-10) codes C18, C19 and 20) between January 2006 and December 2018. Approval for the study was granted by the Public Benefit and Privacy Panel (PBPP) for health and social care, project number 1718-0026. The study meets the requirements set out by the East of Scotland NHS Research Ethics Service for the analysis of secondary National Services of Scotland (NSS) data [25]. This study follows the REporting of studies Conducted using Observational Routinely collected health Data (RECORD) checklist [26].

The final study population included 44,875 patients and the early-stage surgically treated cohort included 24,383 patients. Patient characteristics are described in Table 2. Overall, 32,691 (73%) of patients had a diagnosis of colon cancer and 12,184 (27%) had a diagnosis of rectal cancer. Of those patients diagnosed with colon cancer, 55% were included in the early-stage surgically treated cohort compared to 53% of those diagnosed with rectal cancer. Supplementary Table A, Table B and Table C display the characteristics of patients by cancer network, prior to the early-stage surgically treated cohort selection.

Table 3 presents survival outcomes for the early-stage surgically treated cohort (n = 24,383) and Figure 2 presents the Kaplan-Meier (KM) survival curves for OS and CRC.

Tables 4 and 5 display the results of Cox proportional hazard models. Adjusted models for all-cause mortality and CRC specific mortality show that females have a significantly better survival compared to males. Survival was inversely associated with deprivation. For both cancers, compared to patients in the most deprived quintile, those in the least deprived quintile have a significantly reduced risk of death. Survival improved over time, with diagnoses in later years associated with better outcomes. Further, increased risk of death was observed as comorbidity increases. The results also show that patients diagnosed with colon cancer in SCAN have significantly increased risk of death from all causes and from CRC relative to patients in WoSCAN.

This paper used a national linked administrative dataset to describe factors associated with survival for patients diagnosed with CRC. We described the demographic, diagnostic and treatment characteristics of patients with colon cancer and rectal cancer. Further, for early-stage patients treated with surgery, we assessed the factors affecting overall and CRC specific survival.

The survival models concur with existing UK and other worldwide evidence in a number of areas. Specifically, Cox-proportional hazard models for an early-stage surgically treated cohort of patients with colon or rectal cancer confirm that, stage at diagnosis strongly influences mortality/survival, with those diagnosed at more advanced disease stage having poorer survival compared to those diagnosed at the earliest stage. The adjusted models for deaths from all causes show that patients diagnosed with stage III colon or rectal cancer have a twofold increase in risk of death when compared to those diagnosed at stage I. For CRC specific deaths, this risk increases to almost sixfold for patients diagnosed with colon cancer and almost fourfold for those diagnosed with rectal cancer. These findings are consistent with published national statistics for colorectal cancer in the UK [29] and in other published works reporting the effect of disease stage in Cox-Proportional Hazards models of CRC patients [30–33].

The models show that in general, survival has improved consistently since 2006-08 for patients diagnosed with either colon or rectal cancer. This concurs with other evidence for Scotland and the rest of the UK [34]. Furthermore, as expected, the models find a higher comorbidity score is associated with significantly poorer OS and CRCS for patients diagnosed with both colon and rectal cancer. This finding is consistent with other evidence for the UK and the rest of the world [35–38].

Despite controlling for underlying patient comorbidity as identified from hospital records, the models suggest significant regional variation in outcomes between the three MCNs for patients diagnosed with colon cancer. In particular, patients diagnosed with colon cancer in SCAN appear to have poorer CRC outcomes compared to patients in WoSCAN, both all cause and CRC mortality specifically. These differences suggest potentially inequitable outcomes for patients in an NHS that is designed to ensure equal access to quality care. Other explanations could be selection into the early-stage surgically treated cohort, which could depend on area specific factors that determine this selection. For example, the underlying health of patients in the area; surgeon dependent influences such as comfort in resecting large tumours; and data management system differences. Supplementary Tables A, B and C display the patient characteristics before selection into the early-stage surgically treated cohort and may indicate where possible selection bias should be considered. The models also account for location via the urban/rural indicator. Previous literature shows mixed evidence on the impact of rurality on survival [8, 38] Our models support the view that rurality has no significant impact on survival. However, this may reflect the greater proportion of older individuals captured in our rural cohort compared to the general population. This finding may also differ for later stage patients, where distance to treatment centres is likely to have a bigger impact for those living in rural locations.

Another concerning finding from our study is that area-based deprivation is inversely associated with survival, even after accounting for clinical factors and patient characteristics. This finding is consistent with previous research carried out in Scotland [7]. The reasons for poorer outcomes in the most deprived areas in Scotland is potentially due to later presentation or delay in treatment, however existing evidence from one MCN finds that there is no association between deprivation and these factors [12]. Other explanations include the possibility of comorbidities present within more deprived populations that we have not been able to capture from hospital admissions data or once again the existence of inequities in access to services within these groups. Further investigation into these differences is warranted. In particular, an investigation into type, length and timing of treatment.

In addition, the models find evidence that women have a significantly lower hazard of death (i.e. better survival) compared to men. This result is consistent with previous evidence from Scotland and around the world [39, 40]. The explanation for these differences remains unclear but might be explained partly by endogenous factors such as genetics or hormones, particularly those present in younger females [40, 41]. A further explanation could be differences in access to or use of treatment; however, other evidence from the UK and the US suggests very little difference in survival between sexes [42, 43].

The main strengths of our analysis are that we have been able to utilise population level data, which provides a complete representation of the characteristics and survival outcomes of those diagnosed with CRC in Scotland. Further, the study period covers an extensive time frame which once again adds to the representation of the study population and gives external validity to the results.

However, there are some important limitations to highlight. Firstly, exploration into survival outcomes on the early-stage, surgically treated cohort may be limited by cohort selection bias and differences in data recording between MCN’s, some of which can be seen in Tables A, B and C in the Appendix. Secondly, the selection of the early-stage surgically treated cohort excludes stage IV patients treated with curative intent; therefore, the survival outcomes reported here might be higher than if those patients were included. Related to this, we have not been able to confirm the reliability of use of the therapy-objectives variable as an exclusion criteria to identify patients treated with non-curative or unknown intent. We have also not considered emergency presentations as a factor influencing survival, which may help in our understanding of differences between MCNs. Moreover, using historical datasets where key information is missing for many patients, such as disease stage, is a significant limitation. Although this is improving, linkage to other datasets including chemotherapeutic, radiation and prescribing datasets, will improve our ability to interpret findings on treatment choices and outcomes. In a similar vein, we have not been able to assess any improvement in SMR06 data quality over time. A recent publication by Public Health Scotland assessed the quality of the Scottish cancer registry in 2020/21 and found an accuracy of 98.85% for colorectal cancer [44]. If the ascertainment of colorectal cancer registrations has changed substantially from 2006, this could influence our results if those missed cases differ substantially from those included. Furthermore, we may not have fully captured patient comorbidity in the models because we only had access to hospital admission records and not to diagnoses made in primary care. Some existing research suggests that other measures of comorbidity are superior to the use of Charlson in the colorectal cancer population [45] and that deriving a Charlson index from administrative data sources may result in the under reporting of comorbidities compared to medical records [46]. Finally, the assumption of proportional hazards is violated in all of our models with the exception of the CRCS model for rectal cancer patients. The consequences of non-proportional hazards means that the power of tests is decreased, resulting in a higher likelihood of finding no significant differences when there are truly differences present. Further, for covariates with hazard ratios that increase over time, the relative risk may be overestimated, while for covariates with converging hazards, the relative risk is underestimated [28]. These consequences should therefore be considered when interpreting the model results.

Notwithstanding these limitations, we have highlighted several areas that warrant further research. Specifically, an exploration into the observed inequity in survival between MCNs for colon cancer patients to uncover whether differences in the underlying health of the population are driving variation in survival, or whether system specific factors like practitioner decision making and data management issues play a role. Key next steps in our analysis will be to investigate these areas further and utilise a number of additional administrative data sets that have been linked to the registry data for the purposes of this project [23]. These include detailed information on chemotherapy prescribing and cancer audit data.

In summary, we have used a nationally linked administrative dataset to retrospectively explore patterns in treatment and outcomes for CRC patients in Scotland. We have contributed to the existing evidence base, which is predominantly focused on a single geographic area for all disease stages, by providing evidence for the whole of Scotland, accounting for important prognostic factors such as disease stage. We have demonstrated that Scotland’s unique data linkage infrastructure can accommodate linkage between demographic records, cancer registry and hospital admissions data, providing a fuller picture of the needs of CRC patients. We have identified a number of areas that require further research including regional and socio-demographic differences in outcomes.